In future versions of boinc, it'd be nice if I could manually specify how many WUs of each project to queue if I was so inclined (rather than just setting priorities).

You have two projects.

BOINC keeps track of how much SETI and Predictor it has done, and concentrates on the one with the most debt (the one that is behind on the specified share).

If you've done "more" Predictor, then BOINC will try to do "more" SETI.

... and if that is the case, it will only download more Predictor when there is no chance of getting more SETI in time.

The set of rules that we want BOINC to follow is complex and contradictory.

We don't want to ever return work late.

We want to do work in the specified ratios.

We never want to run out.

---

This kind of phenomena usually has to do with the strategy you decide on to impliment a backup project.

For example, the current LTD values, resource share, queued work deadlines, etc. for the projects onboard are considered when sending work requests to the project schedulers.

Frequently this results is some seemingly strange behaviour when the backup project kicks in, even on always connected hosts.

<edit> BTW, I have observed the grabbing a lot more work than seems necessary from the backup when it kicks in, but I never came up with an explanation for it I *really* liked. Especially since it doesn't appear to happen the same way every time! ;-)

In any event, BOINC managed to work it's way out of what looked like what was going to be a jam when it happened.

HTH,

Alinator

The recent problems with the S@H servers resulted in me draining my S@H workunits. Which, in and of itself isn't really a problem, seeing as my computer is attached to another project (pred@home). However, my computer didn't even bother trying to download new workunits until the last S@H workunit was 100% complete (even though any attempts to download more S@H returned "no work from project"). It was fortunate that my computer happened to be online right when I finished that last S@H workunit, especially since it's only online for about 10 minutes 0-2 times per day. Shouldn't BOINC (or Trux's 36 modification) have downloaded some predictor@home WU before the critical time to guarantee no wasted clock cycles? I think the idea of BOINC is cool - when you run out of work on one project, you still contribute to science, but if I hadn't been online, it wouldn't have gotten any work to process. Now it has about 25 pred@home units to process (which seems to be quite a bit more than I need, especially if S@h comes back online soon).

In future versions of boinc, it'd be nice if I could manually specify how many WUs of each project to queue if I was so inclined (rather than just setting priorities).

---